Focusing the search: A progressively shrinking memetic computing framework

An extremely natural, yet efficient design pattern in memetic computing optimisation is the sequential structure algorithms composed of few simple memes executed sequentially, each one with its own specific role, have proven to be robust and versatile on various optimisation problems with diverse features and dimensionality values. This principle of non-complexity, which can be seen as an application of the Ockham's Razor in memetic computing, leads us to create shrinking three-stage optimal memetic exploration (S-3SOME), a scheme which progressively perturbs a candidate solution by alternating three search operators, the first one being a stochastic global search, the second a random sampling within progressive narrowing hyper-volume, and the third a deterministic local search. Numerical results show that the proposed S-3SOME, despite its simplicity, is competitive not only with other memory-saving schemes recently proposed in literature, but also with complex state-of-the-art populat...

An extremely natural yet efficient design pattern in memetic computing optimisation is the sequential structure algorithms composed of few simple memes executed sequentially, each one with its own specific role, have proven to be robust and versatile on various optimisation problems with diverse features and dimensionality values. This principle of non-complexity, which can be seen as an application of the Ockham's Razor in memetic computing, leads us to create shrinking three-stage optimal memetic exploration (S-3SOME), a scheme which progressively perturbs a candidate solution by alternating three search operators, the first one being a stochastic global search, the second a random sampling within progressive narrowing hyper-volume, and the third a deterministic local search. Numerical results show that the proposed S-3SOME, despite its simplicity, is competitive not only with other memory-saving schemes recently proposed in literature, but also with complex state-of-the-art population-based algorithms characterised by high computational overhead and memory employment.